Compressed recorded video is to be transmitted over a variable bit-rate link, using bandwidth reservation. In order to determining the transmission rate that is to be used (and reserved) at any given time, in such a way as to make the two match so as to minimize wasted bandwidth, the data stream is split into chunks chosen such that the average bit rate for the chunk as a whole is no less that the average bit rate for any shorter chunk starting at the same point. The chunk can then be transmitted using this average rate without any buffering problems. Preferably one chooses chunks so that the average bit rate for the chunk is no less than that for any shorter or longer chunk starting at the same point. This has the benefit that a request for bandwidth never has to ask for an allocation that is higher than that specified in any previous such request. In systems that switch between streams of different degrees compression. the switching points can with advantage be chosen to coincide with the inter-chunk boundaries.
|
1. An automated machine implemented method of transmitting over a communications network a digital data sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, comprising using a programmed processor to:
(a) partition the data sequence into segments;
(b) determine a transmission bit rate for each segment over said communication network; and
(c) transmit the sequence of video signals at a plurality of the determined transmission bit rates,
wherein the sequence is partitioned into segments such that the first segment is that portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible shorter portion starting at the beginning of the sequence, and
wherein each succeeding segment is a portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible shorter portion immediately following the preceding segment.
2. An automated machine-implemented method of transmitting a digital sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, wherein the source video had been coded into a first sequence and a second sequence having respective different compression rates, said method comprising using a programmed processor to:
(a) analyze at least one of the streams to divide it into segments;
(b) select a switching point in the vicinity of an intersegment transition identified at step (a);
(c) if the first sequence was not analyzed in step (a), analyzed the first sequence to divide it into segments;
(d) determine a bit rate for the or each segment of the first sequence up to the switching point;
(e) transmit the signal of the first sequence up to the switching point at the determined bit rate(s);
(f) analyze a modified sequence which includes the second sequence from the switching point onwards, to divide it into segments;
(g) determine a bit rate for segments of the modified sequence; and
(h) transmit the signals of the modified sequence at the determined bit rate(s);
wherein said analyses are each performed by dividing the relevant sequence into segments, wherein the first segment is a portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion, and wherein each succeeding segment is a portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion.
17. An automated machine-implemented method of transmitting a digital sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, comprising using a programmed processor to:
(a) divide the sequence into segments, wherein the first segment is a portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion, and wherein each succeeding segment is a portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion;
(b) determine a bit rate for each segment;
(c) transmit the signals at the determined bit rates
wherein the first segment of the or a sequence is that portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible such portion, and wherein each succeeding segment is that portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible such portion;
in which the bit rate determined for each of at least the later segments of the or a sequence is a number of bits per frame period equal to the lower of:
(i) the highest one of a set of permitted bit rates that is less than or equal to a nominal rate for that segment, said nominal rate being the average coded bits per frame for that segment plus any increase necessitated as a consequence of the determined bit rate for the following sequence being less than the nominal rate for that preceding segment: and
(ii) the highest one of the set of permitted bit rates that is less than or equal to the average coded bits per frame for the preceding segment.
16. An automated machine-implemented method of transmitting a digital sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, comprising using a programmed processor to:
(a) divide the sequence into segments, wherein the first segment is a portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion, and wherein each succeeding segment is a portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion;
(b) determine a bit rate for each segment;
(c) transmit the signals at the determined bit rates
wherein the first segment of the or a sequence is that portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible such portion, and wherein each succeeding segment is that portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any possible such portion
in which the bit rate determined for each of at least the later segments of the or a sequence is a number of bits per frame period equal to the larger of:
(i) the lowest one of a set of permitted bit rates that is greater than or equal to a nominal rate for that segment, said nominal rate being the average coded bits per frame for that segment less any reduction permitted as a consequence of the determined bit rate for the preceding sequence being in excess of the nominal rate for that preceding segment; and
(ii) the lowest one of the set of permitted bit rates that is greater than or equal to the average coded bits per frame for the following segment.
3. A method according to
4. A method according to
5. A method according to
6. A method according to
7. A method according to
8. A method according to
9. A method according to
10. A method according to
11. A method according to
12. A method according to
13. A method according to
14. A method according to
15. A method according to
18. A method according to
19. A method as claimed in
20. A method as claimed in
21. A method as claimed in
|
This application is the US national phase of international application PCT/GB2003/004996 filed 18 Nov. 2003 which designated the U.S. and claims benefit of GB0226872.0 and GB 0322767.5, dated 18 Nov. 2002 and 29 Sep. 2003, respectively, the entire content of each of which is hereby incorporated by reference.
1. Technical Field of Present Example Embodiments
The present invention is concerned with the transmission of digitally coded video signals, for example over a telecommunications network, and, more particularly, video signals which have been encoded using a compression algorithm.
2. Description of Related Art
The rationale of compression algorithms is to exploit the inherent redundancy of the original video signal so as to reduce the number of bits that require to be transmitted. Many such algorithms are defined in international standards such as the ITU H.263 and the ISO MPEG standards. A useful review of these is given in Ghanbari, M., Video Coding, an introduction to standard codecs, IEE, London, 1999.
The degree of redundancy naturally varies with the picture content, and consequently the compression efficiency does too, resulting in a varying number of coded bits per frame. One option is to transmit the bits as they are generated, as in so-called variable bit rate (VBR) systems in which the transmitted bit rate varies considerably with time. Another option—constant bit rate (CBR) systems—is to employ a buffer at both the transmitter and receiver, to smooth out these fluctuations, and transmit from the transmit buffer to the receive buffer at a constant rate. CBR systems utilise a feedback mechanism to vary the rate at which data are generated (for example by adjusting the coarseness of quantisation used, or frame dropping), to prevent buffer overflow. The use of buffering necessarily involves the introduction of delay, increasing the latency of start (LOS)—i.e. the user has to wait while the receive buffer is filled to the necessary level before decoding and display of the pictures can commence. The feedback mechanism involves reduction in picture quality.
It has also been proposed to employ a degree of buffering to reduce, yet not totally eliminate, bit-rate variations (see, for example, Furini, M. and Towsley, D. F., “Real-Time Traffic transmissions over the Internet”, IEEE Transactions on Multimedia, Vol. 3, No. 1, March 2001).
A major consideration when transmitting over a telecommunications network, and in particular packet networks such as the internet, is the effect of network congestion, where packet loss and unpredictable delays can cause problems. This has given rise to proposals for reservation systems, where a transmitter can request the network to allocate a specified guaranteed bit rate for its transmissions for a period of time. One such system, called “RSVP” is described in the Internet Engineering Task Force (IETF) document RFC 2205. However, other systems such as Expedited Forwarding of Differentiated Service, or CR-LDP may also be used.
In the case of a live video feed, the future characteristics of the bitstream being coded are unknown; with recorded material, however, they are. The fact that reservation systems allow the amount of the reserved bit-rate to be changed offers the opportunity to decide on a policy of how much network capacity to reserve at any time, based on knowledge of the coded material. A simple approach is to calculate the peak (VBR, unbuffered) bit-rate and request this for the entire duration of the transmission, but this is wasteful of network capacity and of course the higher the capacity requested, the greater is the probability of the network being unable to provide it and hence of the reservation request being refused. Another simple approach, which minimises the bit-rate to be requested, is to calculate the average bit-rate of the whole transmission and request this; however this results in the need for a very large buffer at the receiver and, more importantly (given that large amounts of storage are today relatively cheap) a large LOS. A modification to the peak-rate approach is considered in the above-cited paper by Furini and Towsley. Their scheme involves identifying the point in the video sequence at which the peak rate reaches a maximum, and requesting this rate for the period of time up to that point. Then the maximum peak rate over the remainder of sequence is located, and this (lower) rate requested. This process continues in the same manner over the whole sequence. The paper also suggests that a degree of buffering might be applied, thereby reducing the effective peak rates before the reservation algorithm is applied. Although this system improves the efficiency of network use as compared with the single peak rate system, there is still much wasted (i.e. reserved but unused) network capacity, and of course the benefit is small if the maximum peak rate occurs towards the end of the sequence. It does however have the benefit that the amount of network capacity requested falls, and, specifically, a reservation request never asks for a bit-rate that exceeds that of the previous requests, thereby reducing the risk of the reservation request being refused.
According to one aspect of the present invention there is provided a method of transmitting a digital sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, comprising:
(a) dividing the sequence into segments, wherein the first segment is a portion at the beginning of the sequence which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion, and wherein each succeeding segment is a portion immediately following the preceding segment which has an average number of coded bits per frame which is greater than or equal to the average number of coded bits per frame of any shorter such portion;
(b) determining a bit rate for each segment;
(c) transmitting the signals at the determined bit rates.
In another aspect, the present invention provides a method of transmitting a digital sequence of video signals which have been encoded using a compression algorithm such that the number of coded bits per frame is not constant, wherein the source video had been coded into a first sequence and a second sequence having respective different compression rates, comprising:
Some embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:
Consider, at a receiver, some arbitrary time segment (but equal to a whole number of frame periods), extending from time tg at which the receiver begins to decode frame g to time th at which the receiver begins to decode frame h. The duration of this segment is h−g. Suppose, further, that the transmission rate during this segment is A bits/frame period.
Obviously, at time tg, the receiver must have already received the bits for all frames up to and including frame g, i.e.
where dj is the number of coded bits generated by the encoder for frame j.
Suppose however that the receiver has, prior to time g, also received p additional bits, that is, in total,
At any time tk (tg≦tk≦th), at which the receiver begins to decode frame k, the receiver has also received (k−g)A bits, so:
Total bits received at time
At this point, the receiver needs to have all the bits for frames up to and including frame k, that is:
Total bits needed at time
Since the number of bits received must be at least equal to the number needed, the condition that needs to be satisfied to avoid buffer underflow is
If this is to be achieved without the transmission of preload bits p, this requires that
Thus, the transmitted rate A must be greater than or equal to the average generated bits per frame over frames g+1 to k, for any value of k (g+1≦k≦h), which will be achieved if
Use of this rate means that the number of bits (h−g)A transmitted during the segment will exceed the number of bits generated for the segment, unless the maximum occurs for k=h, that is, at the end of the segment. On the premise that the continued use of the transmission rate thus calculated, after the maximum has passed, seems to represent the use of a rate higher than absolutely necessary, the first version of the invention now to be described aims to partition the data to be transmitted into segments in such a manner that these maxima always occur at the end of segment.
The first method to be described is for the transmission of stored video material, already encoded using a compression algorithm such as MPEG, over a packet network such as the internet. It presupposes that the network has provision for reservation of bit-rate capacity. It aims to determine the bit-rate that is to be used as a function of time, in such a manner as to achieve:
In this example it is assumed that there is no constraint on the bit rate that may be chosen, and that the bit rate used for transmission and the bit rate reserved on the network are the same.
This first version also is subject to the constraint that the requested bit-rate can never increase—i.e. it is a monotonically decreasing function of time: as noted above, this is desirable in reducing the risk of reservation failure.
Since huge storage hardware is not a problem for current users, in this solution, reducing the required buffer size at the decoder is not of primary concern, though, in fact, the required buffer size resulting from this method is also greatly reduced compared with using the average bit rate to achieve VBR video transmission. Even in the worst case, rarely encountered in practice, the buffer size required would be no larger than required when transmitting a VBR video stream at the average bit rate.
The following algorithm determines a “function of transmission (“FOT”) to be used.
We assume there are N frames in the video sequence, and the number of encoded bits for each frame is d0, d1, . . . dN-1, respectively
As noted above, this algorithm is subject to the constraint that the function of transmission can never increase, but only decrease.
Conceptually, transmission rate changes can occur at any frame interval in the FOT. In practice, there may be a limit on how often the rate may be changed, depending on the constraints of the particular reservation system in use; however, with a monotonically decreasing FOT, a delay in rate change (although wasteful of network capacity), will not result in any loss of quality since its only effect is the reservation of more capacity than is actually needed. The first step of the algorithm is to find how many “steps” the FOT will have, and when each step occurs.
First, we define:
After finding the first “step”, frame (k0+1) is regarded as the “first” frame for the following frames and Ai+1(1) are calculated for i=k0+1, k0+2, . . . N−1. The formula for this is
or, in the general case,
Again, the largest value is chosen as the second “step” edge, at the end of frame k1, k1 being the corresponding value of i. The above procedure is repeated until the last “step” edge at frame N−1 has been reached. In general this results in M values km, m=0, . . . M−1 (where kM-1 is always equal to N−1) which may be regarded as dividing the video sequence into M−1 segments: segment 0 comprises frames 0 to k0; other segments m each comprise frames km−1+1 to km.
The purpose of the second stage of the algorithm is to choose an appropriate transmission rate for the “level” of each “step”. Now, theoretically, the lowest rate that can ensure that all needed bits will be delivered by the end of each “step” even without any preloaded bits is the average of the bit-rates of the frames making up the segment. A lower rate necessarily requires preloaded bits and consequently a higher LOS, whilst with higher rates, network capacity may be wasted. Also, higher rates must lead to more risk of failure to reserve resource.
There are M segments m=0, 1, . . . M−1. Also, we define:
In this case the required rates are simply the average rates Ri=Si/Ki; i=1, 2, . . . M−1.
This method can also be used to calculate the rate R0 for Segment 0, if we define k−1=−1.
Note that, in MPEG video coding, the first frame is always an I frame and it generates more bits than P or B frames. So, often, our results show the first segment includes only one frame and the transmission rate R0 is much larger than R1. Since users easily can wait a few frames' interval to have a better chance of resource reservation success, we prefer to set R0=R1.
The Third Step: After we have determined the whole FOT, the required buffer size at the decoder can be determined.
We now describe a second, modified version which is subject to a constraint on the rates that may be chosen. For example the constraint may be that the rate must be an integer number of bits per frame, or more generally that the rate must be one of a number of discrete rates. In analysis we will use the quantisation operators defined as follows:
Q+(X) means the lowest permitted rate greater than or equal to X (also referred to as the “ceiling” rate);
Q−(X) means the highest permitted rate less than or equal to X (also referred to as the “floor” rate).
Two options will be discussed:
Consider first the ceiling option. We first define the ceiling value of the “height” of the first “step” in the original FOT as the “height” of the refined first “step” in our new FOT. It will be noticed that, in this way, after the first “step”, more bits have been transmitted to the receiver than the sum of bits of frames belonging to the “first” step. Thus, when we refine the second “step”, we should exclude the number of bit belonging to the following “step” but having been transmitted in the previous “step(s)” and recalculate the average rate of the second “step”. If the ceiling value of the new average bit rate is not less than the ceiling value of the average rate of the old third “step”, it is just defined it as the “height” of the refined second “step”. Otherwise, we define the ceiling value of the average bit rate of the old third “step” as the “height” of the refined second “step”. Follow this procedure until the “height” of the refined last “step” is fixed. Since it always takes the ceiling value of each “step”, it is possible that the VBR video stream transmission is achieved a few frame intervals shorter than the duration of the video sequence. With simulating the transmission based on the new FOT, the lifetime duration of FOT can be exactly specified. Once the VBR video stream transmission is achieved, reserved network resource can be immediately released. Thus, the 100% bandwidth utilization is still guaranteed. With the “height” of the refined first “step”, LOS can be precisely recalculated. Finally, through simulating the transmission procedure, the required buffer size to prevent overflow can be also fixed.
The procedure adopted is as follows. Division into segments proceeds as before.
As well as the quantities Si, Ri, Ki, defined above, we also introduce R1i, a temporary value for the transmission rate in Segment i.
I. Calculate all the average rates R1i=Si/Ki; 1=0, 1, . . . M−1.
II. Set the rate for Segment 0 as R0=Q+(R10)
(Note that if as discussed earlier it is desired to use a lower rate for the first Segment, then one may instead begin with Segment 1)
III. Set the rate for Segment 1 by subtracting, before quantisation, the extra bits sent during the previous Segment:
R1=Q+{R11−(R0−R01)}
or =Q+{R21}
whichever is the greater.
IV. For the remaining Segments i=2, . . . M−1:
Ri=Q+{Ri1−(Ri−1−Ri−11)}
or =Q+{Ri+11}
whichever is the greater. Naturally the second alternative does not arise for i=M−1.
The third version to be described uses the “floor” rates. In this case, the processing must be performed in reverse order, starting from the last “step”. This is necessary so that the bits which cannot be transmitted in a particular segment can be pre-transmitted in the previous Segments. The specific procedure first defines the floor value of the average bit rate of the last “step” as the new transmission rate of the refined last “step” in the new FOT. The number of bits which are needed by the refined last “step” but cannot be transmitted, can then be determined. The previous “step(s)” should guarantee such a number of extra bits transmitted before the new last “step” FOT starts. Thus, when we refine the penultimate “step”, we must aim for it to carry the bits it needs itself, plus the extra number of bits needed by the last “step”. So, a new average bit rate has to be re calculated for the second last “step”. If the floor of the new average bit rate of the second last “step” is not bigger than the floor of the average rate of the third last “step” in original FOT, just define it as the “height” of the new second last “step”. Otherwise, define the floor of the average bit rate of the old third last “step” as the “height” of the new second last “step”. Following this procedure until the first “step”, the refinement is achieved and the refined FOT is obtained. As in the ceiling case: with the number of the pre-fetched bits and the “height” of the refined first “step”, LOS can be precisely recalculated; finally, through simulating the transmission procedure, the required buffer size to prevent overflow can be also fixed.
As before, there are M segments m=0, 1, . . . M−1. Also, we define:
I. Calculate all the average rates R1i=Si/Ki; i=1, 2, . . . M−1.
II. Set the transmission rate RM-1 for Segment M−1 equal to the floor value of the average rate for this segment. That is
RM-1=Q−{R1M-1}.
II. Compute the number of preloaded bits, PM-1, that are needed to be present in the receiver buffer at the beginning of Segment M−1 to prevent underflow in Segment M−1.
PM-1=(R1M-1−RM-1)*KM-1
III. The rate for the next segment can then be calculated as
RM-2=Q−{RM-11+PM-1}
or =Q−{RM-31}
IV. This process is then repeated using the general formula, m=M−3, . . . , 0:
Rm=Q−{Rm1+Pm+1}
or =Q−{Rm−11}
whichever is the lower.
And
Pm=R1m−Rm.
Again, if desired this iteration may be stopped at m=1 and R1 used for Segment 0.
This process results in a value for P0 which is a preload for the first Segment, and will need to be transmitted first. In fact, it is convenient to define a preload b0 which includes all bits that are transmitted before the receiver starts decoding the first frame at t=0.
Assuming that R0 is calculated as above, than
b0=P0+R0
If, however the rate R1 is used for Segment 0, then only (K0−1)R1 bits can be transmitted between t=0 and the end of the segment and therefore the total preload is:
P0+K0R0−(K0−1)R1
The latency of start (LOS) is, assuming R1 is used, b0/R1.
The question of buffer size will now be discussed. Certainly, with our FOT, we can get reasonable transmission rate and LOS. The network transmission efficiency can be almost 100%, and it requires a smaller buffer size than the directly using fixed average rate bandwidth. However, in some situations, it is still much bigger than it required by reserving the peak rate bandwidth. In the scheme of reserving the peak rate bandwidth, it is enough if the buffer size at the decoder is only as much as the number of bits spent on the most complicated frame. However, in our scheme, we do need a larger buffer than that. Although, compared with a constant average bit rate, our scheme can get a much smaller buffer size in most situations, it should be admitted that, in the worst situation, the buffer size required by our scheme is near the buffer size required by a constant average bit rate. Such a situation happens when the biggest Ai appears in the last frames of the video sequence. In such a situation, our “down stairs” curve almost has only one “step”. Thus, it would be not effective enough to minimise the buffer size through the “step” changes. Nevertheless, such a situation hardly appears because, the later the “peak bits” appears, the less effect on Ai. Unless, at the end of the sequence, quite a few exceptional complicated frames appear abnormally, it will never happen. No matter what situation happens, LOS will never be a problem with our scheme. We believe, currently, it should not be a problem for the user to have some hardware that has a little large storage. A small LOS and good network transmission efficiency should cause more concern by the users.
In addition, even if users cannot afford a large buffer size our scheme requires, a compromise may be taken between the transmission efficiency and required buffer size at the decoder. With such a compromise, the required buffer size can be further reduced as the users wish.
Incidentally, although our current algorithm description is only based on bits per frame as a basic unit, naturally, the unit can be defined as a GOP or certain number of pictures or packets together. No matter what unit we would like to define in this algorithm, the principle is general and should be in common.
We now describe some examples of coding test video sequences, using the “floor” method. In each case the values of the function of transmission f(t) (or Ri), the value of b0, and a suggested rate for transmission of b0 are given, (a) for the above algorithm (b) using the method of Furini and Towsley, and (c) using a single, average bit rate.
(a) We have coded a test sequence (named Jacknbox) of common intermediate format (CIF), 140 frames in duration with a fixed quantiser of step size 16 using H.263+ and derived the FOT function with our algorithm.
f(t)=
5100
0 < t <= T48;
3645
T48 < t <= T51;
3058
T51 < t <= T52;
2830
T52 < t <= T61;
2682
T61 < t <= T70;
2651
T70 < t <= T71;
2464
T71 < t <= T90;
2447
T90 < t <= T108;
2321
T108 < t.
We define the measure unit of all measurement rates in this document as bits per frame interval.
b0=39824 bits; suggested transmission rate for b0: 5100 bits per frame interval
(b) Using Furini and Towsley's method, we get
f(t)=
9896
T0 < t <= T29;
9432
T29 < t <= T40;
7272
T40 < t <= T41;
6552
T41 < t <= T46;
6184
T46 < t <= T47;
5328
T47 < t <= T48;
3696
T48 < t <= T51;
3632
T51 < t <= T106;
3552
T106 < t <= T138;
2896
T138 < t.
b0=39824 bits
In their transmission scheme, b0 would be achieved by 39824 bits per frame interval.
(c) With a constant average bit rate, the function would be:
f(t)=3669.
b0=108488 bits;
b0 is achieved by 3669 bits per frame interval.
The analysis results are listed in Table 1:
TABLE 1
JacknBox 140 frame, H.263+
(c) Fixed
bandwidth
channel with
(b) Furini &
Schemes
average bit rate
(a) Our scheme
Towsley
Bandwidth
100
100
63.46
utilization
(%)
The Start
3,669
5,100
39,824
Reservation
Rate
(bits/per
frame
interval)
LOS(frames)
108488/3669 = 29.57
39824/5100 =
39824/39824 = 1
7.8
Buffer
108,488
60,336
39,824
Size(bits)
We have also encoded the same video sequence with a CBR rate control. In this case, the LOS would be 29656/3735=7.94 frames. However, 10 frames are skipped with the normal CBR rate control and the bits budget we give is the same as the average number of bits in VBR encoding.
This test used a normal TV programme QCIF (quarter-CIF) sequence with 8400 frames, coded with a fixed Quantiser step size of 16 using H.263+. The picture type is IPPPP . . . with forced updating every 132 frames in the H.263+ recommendation.
(a) f(t)=
4977
T0 < t <= T3173;
4218
T3173 < t <= T3679;
3968
T3679 < t <= T3680;
3848
T3680 < t <= T3681;
3844
T3681 < t <= T4752;
3090
T4752 < t <= T8392;
992
T8392 < t <= T8393;
816
T8393 < t <= T8394;
644
T8394 < t <= T8396;
544
T8396 < t <= T8397;
384
t > T8397;
b0=13944 bits;
As previously, b0 may be achieved by the first rate of 4977 bits per frame interval.
(b) f(x)=
27672
T0 < t <= T8339;
26704
T8339 < t <= T8340;
26560
T8340 < t <= T8341;
26488
T8341 < t <= T8342;
26240
T8342 < t <= T8344;
25832
T8344 < t <= T8345;
25136
T8345 < t <= T8346;
24168
T8346 < t <= T8347;
23816
T8347 < t <= T8352;
23760
T8352 < t <= T8353;
23616
T8353 < t <= T8356;
22824
T8356 < t <= T8357;
22528
T8357 < t <= T8358;
21952
T8358 < t <= T8359;
21744
T8359 < t <= T8369;
20448
T8369 < t <= T8373;
20344
T8373 < t <= T8384;
19960
T8384 < t <= T8385;
19016
T8385 < t <= T8391;
11656
T8391 < t <= T8392;
992
T8392 < t <= T8393;
816
T8393 < t <= T8394;
648
T8394 < t <= T8396;
544
T8396 < t <= T8397;
384
T8397 < t <= T8399.
b0=13944 bits;
b0 may be transmitted at 29762 bits per frame interval.
(c) With a constant average bit rate, FOT would be:
f(t)=3966.
b0=3348584 bits;
b0 may be set at 3669 bits per frame interval.
The analysis results are listed in Table 2:
TABLE 2
8400 frames H.263+
(c) Fixed bandwidth
channel with
(a) Our
(b) Furini &
Schemes
average bit rate
scheme
Towsley
Bandwidth
100
100
14.36
utilization
(%)
The Start
3,966
4,977
27,672
Reservation
Rate
(bits/per
frame
interval)
LOS
3348584/3966 =
13944/4977 =
13944/27672 = 0.5
(frame
844.322
2.8
interval)
Buffer Size
6,116,362
3,908,218
27,672
(bits)
The same TV programme QCIF sequence of 8400 frames was coded using MPEG4, with a fixed Quantiser step size of 10. The picture type is IBBPBBPBBPBB (N=12, M=3). It should be noted that, with B pictures, the encoding sequence of pictures is different from the displaying sequence of pictures. So the related I or P pictures must be transmitted prior to the B picture.
Some pre-processing is needed before using our algorithm.
(a) Finally, our FOT is:
7426
T0 < t <= T4750;
6938
T4750 < t <= T4786;
66470
T4786 < t <= T4798;
6309
T4798 < t <= T4870;
6190
T4870 < t <= T4900;
6083
T4900 < t <= T4918;
6026
T4918 < t <= T8398;
168
T8398 < t.
b0=16548 bits;
b0 can be sent using 7426 bits per frame interval.
(b) f(x)=
57472
T0 < t <= T8338;
50616
T8338 < t <= T8350;
49504
T8350 < t <= T8368;
48608
T8368 < t <= T8371;
48536
T8371 < t <= T8383;
44968
T8383 < t <= T8386;
31752
T8386 < t <= T8389;
28696
T8389 < t <= T8398;
168
T8398 < t.
b0=16040 bits;
b0 may be set at 57472 bits per frame interval.
(c) With a constant average bit rate, FOT would be:
f(x)=6825.
b0=2874758 bits;
b0 may be set at 6825 bits per frame interval.
The analysis results are listed in Table 3:
TABLE 3
8400 frames, MPEG4
(c) Fixed bandwidth
channel with
average
(a) Our
(b) Furini &
Schemes
bit rate
scheme
Towsley
Bandwidth
100
100
11.897
utilization
(%)
The Start
6,825
7,426
57,472
Reservation
Rate (bits/per
frame
interval)
LOS (frames)
2874758/6825 =
16548/7426 =
16040/57472 =
421.21
2.228
0.279
Buffer Size
6,236,252
3,997,072
57,472
(bits)
From the above experimental results, it can be seen that LOS has been greatly reduced while we still keep 100% transmission efficiency. No network resource has been wasted. The only thing still can be further improved is to further minimise the required buffer size at the decoder.
Also stored in the disc store 12 is a computer program 21 for implementing controlling the operation of the server. The operation of this program, using the “floor” method, will now be described with reference to the flowchart shown in
Step 100
A request is received via the interface 15 from a remote terminal for transmission of a desired video sequence; such a request includes the filename of that one of the files 20 containing that sequence.
Step 101
The processor 10 reads the file in question from the disc store 12 and determines the number of coded bits dj in the file for each of the N frames in the stored sequence, and stores the value of N and dj (j=0 . . . N−1) in the memory 11.
Step 102
The processor calculates kO . . . kM-1 as described above and stores M and kO . . . kM-1 in the memory 11.
Step 103 Calculate R1i=for all i
Step 104 Set RM-1=Q
Step 106 Set a pointer m=M−2
Step 107 Compute Rm and Pm
Step 109 Decrement m. If m≧0, go to step 107
Step 111 Compute b0=P0+R0
Step 112 Compute the segment durations—in this implementation the preload and segment 0 are regarded as a single segment for transmission purposes. Thus
τ0=(b0/R0+k0+1)*τ
τi=(ki−ki−1)*τ i=1 . . . M−1
where τ is the length of a frame period.
Step 113 Set i to 0
Step 114 Transmit a reservation request specifying a rate of Ri and a duration of at least τi.
Step 115 Transmit Segment i at rate Ri (preceded, when i=0, by P0 preload bits).
Step 116. If all segments have been transmitted, stop; otherwise, increment i at 117 and go to step 114.
Some reservation systems, such as the RSVP system mentioned earlier, in order to accommodate multicasting, require that a reservation request be issued by the receiving terminal. In such a case step 113 would be modified to specify the transmission of a message to the receiving terminal specifying Ri and τi. Whereupon the terminal would transmit the required reservation request to the network.
In some networks, it may be that there is some constraint on the times at which the reserved rate may be changed. However the approach adopted above is robust to such problems because every reservation request except the first requests a lower rate than before. It follows that delay in processing such requests results in the reserved rate remaining high after the actual transmitting rate has been reduced. In this case the efficiency of network utilisation falls, but the transmission quality is unaffected.
The reservation algorithm described above is built upon the constraint that the reserved bit rate must never be increased. This is not however essential, so a second embodiment of the invention which is not subject to this constraint will now be described.
In the case, each segment is chosen in such a manner that, as before, the average generated bit rate Σdj for the segment is greater than or equal to the average for any shorter part of the video sequence beginning at the start of that segment, but, now, may be less than the average for some longer part starting at the same point.
The procedure will be described for the general segment q (=0 . . . M−1)
Using
calculate Ai(q) for all kq-1+1≦i≦kq-1+H (or kq-1+1≦i≦N−1 if this is shorter)
where H is some defined maximum length that is to be permitted.
Find the value of i for which Ai(q) is largest, and set kq equal to this value of i.
This is the same as the previously described procedure, except that the search for the maximum average rate is restricted in its range.
Once kq (q=0, . . . , M−1) have been determined, the actual transmission rates can be determined exactly as described above except that any limits defined to prevent a rate from exceeding that of the preceding segment, or from falling below that of the following one, are omitted.
A second embodiment of the invention explores the possibility of video rate switching. Here, two (or more) video streams are generated, with different picture qualities and hence different data rates. Typically, these may be generated by using different coarsenesses of quantisation—i.e. the low-quality, low data rate stream uses a coarse quantiser and a higher-quality stream, having a higher data rate, uses a less coarse quantiser.
The possibility of video rate-switching is of particular interest in the present context where perhaps rate reservation failure occurs at the beginning of a transmission, and the situation can be remedied by firstly sending a relatively poor quality stream, and later switching to a higher quality stream when the nature of the signal and/or network conditions allow it. However, the system to be described is also useful where video rate-switching is used for some other reason.
When inter-frame coding is in use, switching between two different streams can cause serious corruption of the picture due to mistracking of predictors at the coder and decoder: however, switching may be accommodated without such degradation of picture quality by generating, from time to time, transitional coded frames which essentially code up the difference between a frame of the stream that is to be switched to and a frame of the stream that is to be switched from. So transmission of frames from a first stream is followed by one of more transitional frames and then by frames from the second stream. The generation of such transitional frames is not new and will not be described further. For a descriptions of such a system, see our international patent application WO 98/26604 (and corresponding U.S. Pat. No. 6,002,440). Another such system, using so-called “SP-frames” is described in Marta Karczewicz and Ragip Kurceren, “A Proposal for SP-frames”, document VCEG-L-27, ITU-T Video Coding Experts Group Meeting, Eibsee, Germany, 9-12 Jan. 2001, and Ragip Kurceren and Marta Karczewicz. “SP-frame demonstrations”, document VCEG-N42, ITU-T Video Coding Experts Group Meeting, Santa Barbara, Calif., USA, 24-27 Sep. 2001.
In the context of the “FOT” approach described above, the question of switching between two streams presents some questions that need to be addressed. If one considers switching at an arbitrary point in time from a first stream to a second stream, then in general the decoder buffer will contain frames of the first stream, which are not useful for decoding the second stream. Thus, assuming that the decoder is to immediately switch to decoding of the second stream, these frames will be unused and represent wasted transmission capacity. Worse, frames needed for decoding of the second stream will not be present in the buffer. Theoretically this can be accommodated if the FOT for the second stream is recalculated, considering the beginning of the part of the second stream that is actually to be transmitted to be the start of the stream, but in practice this can result in a prohibitively high transmitted data rate requirement if interruption of the displayed pictures is to be avoided.
The problem of wasted bits can be avoided by allowing the decoder to continue to decode the frames of the first stream that remain in the buffer, and during this period the buffer might accumulate some of the frames that are needed for decoding of the second stream (i.e. transitional frame(s) and frames of the second stream) but nevertheless the danger of an excessive transmitted bit-rate requirement remains.
Ideally, bitstream switching should occur as soon as available bandwidth appears. However, owing to the problems just discussed, this is not practical. Also, if transitional frames—which normally are generated only at selected points rather than for every frame—are to be generated then the points (the switching points) at which these are to be provided should preferably be planned in advance.
Based on such considerations, we first consider the possibility of switching at times which coincide with the “edge” of a “step” of the FOT. It is a characteristic of this scheme that, at the “edge” of each “step”, the receiver buffer stores no bits, as all transmitted bits have been decoded into pictures. Thus, if one were to switch at the “edge” of the original stream, all transmitted bits would be emptied from the receiver buffer and no bits would be wasted due to the bitstream switching.
Although setting switching points at the “step edges” of the original bit stream may waste no transmitted bits, there would be still a problem if the switching point in the new stream were not at a “step edge”. The reason is that if the switching point is not at a “step edge” in the new stream, some pre-accumulated bits for the new stream might have to be transmitted within a very short space of time in order to play video continuously at the receiver. It might lead a much higher rate reservation request, perhaps even higher than the reservation rate that the new stream implies. If the switching point in the new bitstream is at the middle of a “step”, the shortage of accumulated bits results in a high rate reservation. Thus, ideally, the switching point in the new video stream should be also at the “step edges”.
According to the above analyses, it might seem that the only chance to have the optimum switching points for the two streams is where they have the same “edge points”. Otherwise, either bits are wasted or one requires a very high bit rate after bitstream switching. Upon further investigation, fortunately, we have found that, for the FOT curves generated from different quantisers, do have similarly positioned “step edges”, even if they are not absolutely the same. The reason is that, in a video sequence, complex pictures must cost more bits than normal ones no matter what quantiser is selected.
We have verified this with some experiments. In the experiment, a 140 CIF Jacknbox video sequence was selected.
In the first experiment, we wish to clarify whether different video streams based on the same video sequence approach their “step edges” together in their FOT. In
In the second experiment, we suppose that we are to switch the bit stream (Q16 stream) generated with a fixed quantiser 16 to a second bit stream (Q8 stream) generated by a fixed quantiser 8 at each frame interval. In
In
As for the question of precisely at what point to choose a switching point for switching from a first to a second stream in practice, if the steps of the two streams coincide, there is of course no ambiguity. If however there is a difference in timing, one can:
However in practice it matters little which option one selects since the differences between them in terms of performance are fairly small: indeed, if the chosen switching point is a few frames offset from the “step”, satisfactory performance can often be obtained.
In the light of this, the proposed method proceeds as follows (assuming option (a) above):
i) Calculate the FOT for the first stream;
ii) Choose a switching point to coincide with a step of this FOT;
iii) Generate a transition frame;
iv) Calculate the FOT for the transition frame plus the remainder of the second stream;
v) Transmit the first stream up to the switching point;
vi) Transmit the transition frame plus the remainder of the second stream.
In the event that option (b), (c) or (d) is used, then step i) would involve calculation of the FOT of the second stream too, and step (ii) would involve selection according to the option chose. Nevertheless, the FOT for the second stream still has to be recalculated in step 4. Note also that the (re)calculation at step (iv) will automatically take into account any corrections necessary due to non-coincidence of the switching point with the step originally calculated for the second stream, and/or due to the use of the “ceiling” or “floor” rates as discussed earlier.
Of course, more than one switching point may be chosen, if desired, for example to revert to the first stream, or to switch to a third stream.
Although the switching issues have been discussed in the context of systems that are constrained to have a monotonically decreasing FOT, this approach to switching may also be used where this constraint is not applied. Equally, it is also useful when switching from a high-quality stream to a low-quality stream.
Patent | Priority | Assignee | Title |
10075744, | Feb 11 2010 | DISH TECHNOLOGIES L L C | Systems and methods to provide trick play during streaming playback |
10116722, | Aug 06 2007 | DISH TECHNOLOGIES L L C | Apparatus, system, and method for multi-bitrate content streaming |
10165034, | Aug 06 2007 | DISH Technologies L.L.C. | Apparatus, system, and method for multi-bitrate content streaming |
10225304, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for adaptive-rate shifting of streaming content |
10298985, | May 11 2015 | MEDIAMELON, INC. | Systems and methods for performing quality based streaming |
10469554, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
10469555, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
10951680, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
11076187, | May 11 2015 | MEDIAMELON, INC. | Systems and methods for performing quality based streaming |
11470138, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
11677798, | Apr 30 2004 | DISH Technologies L.L.C. | Apparatus, system, and method for multi-bitrate content streaming |
11991234, | Apr 30 2004 | DISH Technologies L.L.C. | Apparatus, system, and method for multi-bitrate content streaming |
8370514, | Apr 28 2005 | DISH TECHNOLOGIES L L C | System and method of minimizing network bandwidth retrieved from an external network |
8402156, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
8612624, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
8683066, | Aug 06 2007 | ECHOSTAR TECHNOLOGIES L L C | Apparatus, system, and method for multi-bitrate content streaming |
8838824, | Mar 16 2009 | DILITHIUM ASSIGNMENT FOR THE BENEFIT OF CREDITORS , LLC; Onmobile Global Limited | Method and apparatus for delivery of adapted media |
8868772, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for adaptive-rate shifting of streaming content |
8880721, | Apr 28 2005 | DISH TECHNOLOGIES L L C | System and method for minimizing network bandwidth retrieved from an external network |
8955024, | Feb 12 2009 | British Telecommunications public limited company | Video streaming |
9060189, | Dec 10 2008 | British Telecommunications public limited company | Multiplexed video streaming |
9071668, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
9160778, | Oct 26 2011 | NOKIA SOLUTIONS AND NETWORKS OY | Signaling enabling status feedback and selection by a network entity of portions of video information to be delivered via wireless transmission to a UE |
9167257, | Mar 11 2008 | British Telecommunications public limited company | Video coding |
9344496, | Apr 28 2005 | DISH TECHNOLOGIES L L C | System and method for minimizing network bandwidth retrieved from an external network |
9407564, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for adaptive-rate shifting of streaming content |
9510029, | Feb 11 2010 | ECHOSTAR TECHNOLOGIES L L C | Systems and methods to provide trick play during streaming playback |
9571551, | Apr 30 2004 | DISH TECHNOLOGIES LLC | Apparatus, system, and method for multi-bitrate content streaming |
Patent | Priority | Assignee | Title |
5612900, | Jun 06 1995 | Kabushiki Kaisha Toshiba | Video encoding method and system which encodes using a rate-quantizer model |
5991811, | Sep 04 1995 | Kabushiki Kaisha Toshiba | Information transmission system utilizing both real-time data transmitted in a normal-in-time direction and in a retrospective-in-time direction |
6002440, | Dec 10 1996 | British Telcommunications public limited company | Video coding |
6522693, | Feb 23 2000 | International Business Machines Corporation | System and method for reencoding segments of buffer constrained video streams |
6937770, | Dec 28 2000 | EMC IP HOLDING COMPANY LLC | Adaptive bit rate control for rate reduction of MPEG coded video |
20010025308, | |||
20020090027, | |||
WO9826604, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Nov 18 2003 | British Telecommunications public limited company | (assignment on the face of the patent) | / | |||
Jan 14 2004 | GHANBARI, MOHAMMED | British Telecommunications public limited company | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 017461 | /0077 | |
Jan 14 2004 | SUN, KAI | British Telecommunications public limited company | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 017461 | /0077 |
Date | Maintenance Fee Events |
Jan 17 2014 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Feb 07 2014 | ASPN: Payor Number Assigned. |
Jan 09 2018 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Dec 15 2021 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
Date | Maintenance Schedule |
Jul 20 2013 | 4 years fee payment window open |
Jan 20 2014 | 6 months grace period start (w surcharge) |
Jul 20 2014 | patent expiry (for year 4) |
Jul 20 2016 | 2 years to revive unintentionally abandoned end. (for year 4) |
Jul 20 2017 | 8 years fee payment window open |
Jan 20 2018 | 6 months grace period start (w surcharge) |
Jul 20 2018 | patent expiry (for year 8) |
Jul 20 2020 | 2 years to revive unintentionally abandoned end. (for year 8) |
Jul 20 2021 | 12 years fee payment window open |
Jan 20 2022 | 6 months grace period start (w surcharge) |
Jul 20 2022 | patent expiry (for year 12) |
Jul 20 2024 | 2 years to revive unintentionally abandoned end. (for year 12) |